wassname/adapters_can_monitor_lies

Stars
2
Language
Jupyter Notebook
Created 6 months ago
Updated 5 months ago

wassname/adapters_can_monitor_lies

wassname

There are no reviews yet. Be the first to send feedback to the community and the maintainers!

inspired by circuit breakers paper. honesty>harmless

rl-portfolio-management

Attempting to replicate "A Deep Reinforcement Learning Framework for the Financial Portfolio Management Problem" https://arxiv.org/abs/1706.10059 (and an openai gym environment)

Jupyter Notebook

keywordshitter2

A website to find long-tail keywords using search suggestions

viz_torch_optim

Videos of deep learning optimizers moving on 3D problem-landscapes

Jupyter Notebook

attentive-neural-processes

implementing "recurrent attentive neural processes" to forecast power usage (w. LSTM baseline, MCDropout)

Jupyter Notebook

world-models-sonic-pytorch

Attempt at reinforcement learning with curiosity for Sonic the Hedgehog games. Number 149 on OpenAI retro contest leaderboard, but more work needed

Jupyter Notebook

satellite_leak_detection

Detect water leaks from satellite images using machine learning

Jupyter Notebook

pipe-segmentation

Using machine learning to find water pipelines in aerial images with 73% accuracy

Jupyter Notebook

phoneme2grapheme

Teaching machines to spell with deep learning (acc=>80%) e.g. a model hears "pɹˈaʊd˺ɚ" and writes "prowder" (but it should be "prouder")

Jupyter Notebook

rl_2d_walker.js

Teaching a humanoid to walk(ish), then displaying in your browser (using tensorflow.js and reinforcement learning)

Jupyter Notebook

compare_altcoin_development

Compare development stats between altcoins

seq2seq-time

Bechmarking seq2seq models on a range of multivariate regression datasets

Jupyter Notebook

valuations_of_ethereum

ascii2segy

This script uses segpy to convert ascii files to segy

awesome-rlhf

Lists of datasets, training, and evals for RLHF and similar

Phonetic_english

Fork of PIE - Phonetically Intuitive English (PIE), to make it compatible with other respelling schemes, primarily Wikipedia's excellent one.

simple_gpt2_chatbot

Jupyter Notebook

svg2cube

Generate isometric game sprites. Inputs an svg panel and it's folded into a cube and rendered from any angle.

awesome-interpretability

Awesome tools for interpreting, manipulating the internals of of deep neural networks.

detect_bs_text

Can we measure how good a text is by how much an LLM learns from it?

Jupyter Notebook

open_pref_eval

Hackable, simple, llm evals on preference datasets

compare_github_repos

Compare github repositories by commits/w, stars, size, contributors, and other advanced statistics.

join-dic-hunspell

Join dic files to make a dic file for hunspell

lie_elicitation_prompts

Research dataset. We use prompts to get LLM's to lie. Using sys prompts and multi shot examples

Jupyter Notebook

LoRA_are_lie_detectors

Experiment to see if low rank adapters can work as interventions for lie detection on LLM's

Jupyter Notebook

train-seismic-interp

Interpret a synthetic seismic line so you can measure and improve your error.

repr-preference-optimization

cardsforscience

cardsforscience.com a solitaire game where you work out the secret rule (inspired by Eleusis)

iris_bigvae

experiment: IRIS but with pretrained LLM

Jupyter Notebook